PeerInfinity comments on Less Wrong: Open Thread, September 2010 - Less Wrong

3 Post author: matt 01 September 2010 01:40AM

You are viewing a comment permalink. View the original post to see all comments and the full post content.

Comments (610)

You are viewing a single comment's thread. Show more comments above.

Comment author: Emile 08 September 2010 10:29:11PM 8 points [-]

Note also the Wiki page, with links to previous threads (I just discovered it, and I don't think I had noticed the previous threads. This one seems better!)

One interesting game topic could be building an AI. Make it look like a nice and cutesy adventure game, with possibly some little puzzles, but once you flip the switch, if you didn't get absolutely everything exactly right, the universe is tiled with paperclips/siny smiley faces/tiny copies of Eliezer Yudkowsky. That's more about SIAI propaganda than rationality though.

One interesting thing would be to exploit the conventions of video games but make actual winning require to see through those conventions. For example, have a score, and certain actions give you points, with nice shiny feedbacks and satisfying "shling!" sounds, but some actions are vitally important but not rewarded by any feedback.

For example (to keep in the "build an AI" example), say you can hire scientists, and the scientists' profile page lists plenty of impressive certifications (stats like "experiment design", "analysis", "public speaking", etc.), and some filler text about what they did their thesis and boring stuff like that (think: stats get big Icons, and are at the top, filler text looks like boring background filler text). And once you hired the scientists, you get various bonuses (money, prestige points, experiments), but the only of those factors that's of any importance at the end of the game is whether the scientist is "not stupid", and the only way to tell that is from various tell-tale signs for "stupid" in the "boring" filler texts - For example things like (also) having a degree in theology, or having published a paper on homeopathy ... stuff that would indeed be a bad sign for a scientist, but that nothing in the game ever tells you is bad.

So basically the idea would be that the rules of the game you're really playing wouldn't be the ones you would think at first glance, which is a pretty good metaphor for real life too.

It needs to be well-designed enough so that it's not "guessing the programmer's password", but that should be possible.

Making a game around experiment design would be interesting too - have some kind of physics / chemistry / biology system that obeys some rules (mostly about transformations, not some "real" physics with motion and collisions etc.), have game mechanics that allow you to do something like experimentation, and have a general context (the feedbacks you get, what other characters say, what you can buy) that points towards a slightly wrong understanding of reality. This is bouncing off Silas' ideas, things that people say are good for you may not really be so, etc.

Here again, you can exploit the conventions of video games to mislead the player. For example, red creatures like eating red things, blue creatures like eating blue things, etc. - but the rule doesn't always hold.

Comment author: PeerInfinity 09 September 2010 05:23:22PM 2 points [-]

"once you flip the switch, if you didn't get absolutely everything exactly right, the universe is tiled with paperclips/tiny smiley faces/tiny copies of Eliezer Yudkowsky."

See also: The Friendly AI Critical Failure Table

And I think all of the other suggestions you made in this comment would make an awesome game! :D

Comment author: Emile 09 September 2010 06:00:20PM 0 points [-]

Ooh, I had forgot about that table - Gurps Friendly AI is also of interest.